AI-Powered “Gn Calculator” Personalizes Gonadotropin Dosing for PCOS—A Game-Changer in IVF, study suggests

How can you confidently individualize gonadotropin starting doses for women with PCOS—balancing oocyte yield, safety, and cost? A new study in BMC Pregnancy and Childbirth introduces a machine learning model that delivers just that, potentially transforming controlled ovarian stimulation protocols.

A New Challenge in PCOS-IVF: Hitting the “Goldilocks Zone”

Women with polycystic ovary syndrome (PCOS) present a paradox during IVF: too high a gonadotropin (Gn) dose risks ovarian hyperstimulation syndrome (OHSS), while too low reduces the chance of a successful cycle. Achieving an oocyte yield in the 7–16 range offers the best compromise for live birth rate, safety, and cost—but how to tailor the starting dose for each patient?

Study Design: Machine Learning Meets Clinical Medicine

Researchers retrospectively analyzed 976 women with PCOS who underwent GnRH-antagonist IVF protocols at two centers. Patients were grouped by oocyte yield: low (<7), optimal (7–16), or high (>16). The team used random forest (RF) and LASSO regression—two machine learning strategies—to identify the most important predictors of ovarian sensitivity (OS), defined as oocytes retrieved per unit Gn starting dose.

Key Predictors: Four Variables Drive the Model

Across all statistical approaches, just four baseline variables—BMI, AMH, basal FSH (bFSH), and antral follicle count (AFC)—emerged as the most powerful predictors of ovarian response. A robust RF model, using only these four, predicted OS with an R² of 0.86 in an independent test set.

Targeted Dosing: From Prediction to Practice

Using the model, clinicians can input a patient’s BMI, AMH, bFSH, and AFC, plus a desired oocyte target, into an online “Gn Calculator.” The algorithm then recommends a starting Gn dose tailored to that individual. Simulations showed the model would have recommended higher doses for low responders and lower doses for hyper-responders, aligning more patients with the optimal 7–16 oocyte window. Dose predictions for optimal responders strongly matched historical expert practice.

Clinical Implications: Safer, More Efficient, More Personalized IVF

Precision dosing supports both safety and efficacy, minimizing OHSS while preventing undertreatment.

The model is easy to use and based on routine labs and ultrasound findings, aiding real-world implementation.

By recommending targeted adjustments, the tool may reduce treatment time and costs, while improving cumulative live birth rates.

Conclusion

Machine learning can now help clinicians individualize Gn starting doses for women with PCOS, enhancing safety, efficiency, and live birth outcomes. The online calculator is ready for clinical use, pending prospective validation.

Key points

A machine learning model using BMI, AMH, bFSH, and AFC can accurately predict ovarian response in PCOS.

The online "Gn Calculator" provides individualized gonadotropin starting doses for GnRH-antagonist IVF cycles.

Targeting an oocyte yield of 7–16 maximizes outcomes and safety for women with PCOS.

Model-guided dosing may reduce both undertreatment and overtreatment, lowering costs and OHSS risk.

Prospective multicenter studies are needed to confirm clinical benefit in real-world practice.

Citation:

Meng W, Pan Y, Zhang M, et al. Estimating individualized gonadotropin starting doses in women with PCOS: a target-driven machine learning model for GnRH-antagonist stimulation. BMC Pregnancy and Childbirth. 2026. https://doi.org/10.1186/s12884-026-09783-x